Franççois Yale a &C laired urand b. M éétropolitain b U niversitéédem ontrééal. Available online: 01 Sep 2011

Transcription

1 This article w as dow nloaded by: [U niversite de M ontreal] On: 19 Septem ber 2011, At: 05:48 Publisher: Routledge Inform a Ltd Registered in England and W ales Registered N um ber: Registered office: M ortim er H ouse, M ortim er Street, London W 1T 3JH, U K Am erican Review of Canadian Studies Publication details, including instructions for authors and subscription inform ation: w w.tandfonline.com /loi/rarc20 What did Quebeckers Want? Impact of Q uestion W ording, Constitutional Proposaland Context on Support for Sovereignty, Franççois Yale a &C laired urand b a A gencedesantééetdesservicessociauxdum ontrééal M éétropolitain b U niversitéédem ontrééal Available online: 01 Sep 2011 To cite this article:franççois Yale & Claire Durand (2011): W hat did Quebeckers W ant? Im pact of Question W ording, ConstitutionalProposaland Context on Support for Sovereignty, , Am erican Review of Canadian Studies, 41:3, To link to this article: PLEASE SCRO LL D O W N FO R ARTICLE Fullterm s and conditions of use: w w.tandfonline.com /page/term s-andconditions This article m ay be used for research, teaching and private study purposes. Any substantialor system atic reproduction, re-distribution, re-selling, loan, sub-licensing, system atic supply or distribution in any form to anyone is expressly forbidden. The publisher does not give any w arranty express or im plied or m ake any representation that the contents w illbe com plete or accurate or up to date. The accuracy of any instructions, form ulae and drug doses should be independently verified w ith prim ary sources. The publisher shallnot be liable for any loss, actions, claim s, proceedings, dem and or costs or dam ages w hatsoever or how soever caused arising directly or indirectly in connection w ith or arising out of the use of this m aterial.

2 American Review of Canadian Studies Vol. 41, No. 3, September 2011, What did Quebeckers Want? Impact of Question Wording, Constitutional Proposal and Context on Support for Sovereignty, François Yale a and Claire Durand b a Agence de Santé et des Services Sociaux du Montréal Métropolitain; b Université de Montréal The idea of sovereignty has been part of Québec s political landscape for more than 45 years, giving rise to two referendums. Throughout this period, pollsters surveyed the population. Using close to 700 polls published since the first election of Parti Québécois in 1976, this article examines how support for sovereignty evolved outside the referendum campaigns and since the last one in the Fall of Using longitudinal multilevel analysis, it takes into account two levels of explanation: first, the wording of the question, which includes the type of constitutional proposal polled; and second, the passage of time and the events that marked it. This meta-analysis confirms that sovereignty combined with an association with the rest of Canada has always stood apart from more extreme constitutional options as having the highest support, and it therefore automatically resurfaced at the start of any referendum campaign. It also confirms that the different events that marked the period only had a temporary effect on support for sovereignty. Keywords: sovereignty; polls; question wording; longitudinal multilevel analysis; Québec The idea of sovereignty has been part of Québec s political landscape for over 45 years, giving rise to two referendums over a period of less than two decades one on May 20, 1980, and another on October 30, Throughout this period, the issues surrounding sovereignty remained at the forefront of the public arena. In such a landscape where the future of a country was to be decided upon, the presence of opinion pollsters was tangible. Although the pollsters mandate was to take the pulse of the population on the sovereignty issue, polls also played a significant role in informing the public of the parties strategies. This was no easy task. Over the years and the ensuing discussions, debates, and political strategies, many different expressions were used to name the nationalist project. The most commonly used were separation, independence, sovereignty, sovereigntyassociation, and sovereignty-partnership. To varying degrees, all the proposals for what we shall refer to as sovereignty now form the symbolic background of the Québec sovereigntist movement. Thus, the challenge of pollsters was to take this reality into account when deciding on their survey questions. However, pollsters also act on their clients requests and ask questions on constitutional proposals that are not in the public debate, pertaining, for example, to separation or independence. This article examines how support for sovereignty has changed since the issue first appeared on Québec s political landscape, based on published, private, and academic polls, taking into account two levels of explanation: first, the wording of the question, including both the wording itself and the ISSN print/issn online 2011ACSUS DOI: /

3 American Review of Canadian Studies 243 constitutional proposal to which it refers, and second, the passage of time and the events that marked it. Historical context The idea of Québec sovereignty really began to take root with the Rassemblement pour l Indépendance Nationale (RIN) in the second half of the 1960s. The term used at the time was independence. The Parti Québécois (PQ), which emerged from the Mouvement Souveraineté-Association (MSA) founded by René Lévesque in 1968, added an associative dimension with the rest of Canada that led to the sovereignty-association project. When the PQ came to power in November 1976, it had promised that a referendum on sovereignty-association would be held at some point during its five-year mandate. However, aware that its project could hardly gain the support of the majority of the population according to the polls published at that time, it made a mandate to negotiate the object of the 1980 referendum. Following the decisive victory by the no side in that referendum (59.6 percent of the vote with a participation rate of 85.6 percent), the PQ decided to move on to other issues what was termed le beau risque or good risk. This period ended with the repatriation of the Canadian Constitution by the Federal Government led by Pierre-Elliott Trudeau in 1982, without Québec s approval. Most of the Québec political elites considered this move as a betrayal. However, the Québec Liberal Party (QLP), led by Robert Bourassa, was elected in December 1985 and, between 1985 and 1988 in view of the dead-end in which federal-provincial politics lay at that time came a period during which the issue of sovereignty ceased to appear in published opinion polls. In 1988, Jacques Parizeau, the former minister of finance of the Lévesque government and considered to be a PQ hardliner on sovereignty, breathed new life into the movement. His more radical position excluded the associative dimension from the sovereignty project. This triggered a new round of constitutional negotiations initiated by the new Canadian Prime Minister Brian Mulroney who had promised during the electoral campaign in 1984 to reintegrate Québec in the Constitution in honor and enthusiasm. Mulroney managed to settle a deal with the 10 Canadian provinces that would satisfy the five main demands of Québec, such as the recognition that Québec formed a distinct society within Canada. This deal, called the Meech Lake Accord, had to be voted in by all 10 provinces within three years in order to be enforced, in accordance with the Constitution Act, 1982 amending formula known as the unanimity procedure. However, during these three years, some provincial governments changed and so did minds. Opposition against the agreement built, including among the First Nations. By June 1990, as the deadline for the acceptance of the agreement was finally up, three provinces were missing, either because they had not taken avoteorbecausetheyhadvotedagainsttheagreement.thiswasgenerallyperceivedin Québec as a lack of flexibility and tolerance on the part of the rest of Canada. The failure of the Meech Lake Accord, however, was followed by another round of constitutional negotiations, which ended with the Charlottetown Accord signed in August This agreement, supported by all Premiers and main political parties, was to be accepted, provided that the majority of Canadians and the majority of residents in each province vote in its favor in a referendum. Three provinces British Columbia, Alberta and Québec had already passed legislation requiring that constitutional amendments be submitted to a public referendum. This gave the impetus to hold a referendum throughout Canada though it was not mandatory. The referendum held on October 26, 1992, ended with a 54.3 percent no throughout Canada and a 56.7 percent no in Québec.

4 244 F. Yale and C. Durand Following these two failures, the Parti Québécois, with Jacques Parizeau as leader, was elected on September 12, 1994, and promised to hold a referendum pertaining to sovereignty itself as soon as possible in the following mandate (Parti Québécois 1994, 16) without any mention of an association with the rest of Canada. This approach was abandoned once again, just before the 1995 referendum campaign due to a lack of popular support, at least according to a reading of the opinion polls of the time. Instead, the October 1995 referendum campaign dealt with sovereignty-partnership, resulting from an agreement between three parties: the PQ and Action Démocratique du Québec at the provincial level, and the Bloc Québécois, a sovereignist party acting at the federal level that had emerged in the run-up to the failure of the Meech Lake Accord. The referendum, held on October 30, 1995, ended with a very close result (50.6 percent no with a participation rate of 93.5 percent). The option proposed then appears to have remained that of the PQ until Since 2006, however, the PQ s constitutional position has been the object of many debates, both within and outside the party. The period since the last referendum has seen numerous changes in governments, both at the federal and Québec levels. However, the event characteristic of that period is the sponsorship scandal which came to the public eye when it was revealed that the Canadian Government, headed by the Canadian Liberal Party (CLP) leader Jean Chrétien, had spent money sponsoring events in Québec in order to increase its visibility. A report of the Auditor General of Canada tabled in February 2004 showed that the sponsorship program had been ill-administered, and that much of the money might not have been properly allocated. Finally, a Commission headed by Judge John Gomery revealed that a number of people had benefitted from the program s subsidies without doing anything for it, and that part of the money had ended up with the Canadian Liberal Party. Multiple approaches, diverse results Sociologists and political scientists have studied popular support for the PQ s sovereignty proposals throughout the history of this issue. What is the proportion of Québeckers that supports Québec sovereignty? What proportion of the electorate will vote for or against sovereignty in a referendum? What are the determinants (sociological, political, economic, psychological, etc.) of opinion, voter intention and, ultimately, the final referendum vote? These are all questions that researchers have, in their own ways, sought to answer. The two most significant theoretical approaches to the issue that can be identified are, the Montréal school (also called the rational choice approach (Blais, Nadeau, and Martin 1995; Blais, Martin, and Nadeau 1998; Martin and Nadeau 2000; Nadeau and Fleury 1995; and Nadeau, Martin, and Blais 1999) and the psycho-sociological school, revolving around the work of Pinard and Hamilton (1986), and Pinard et al. (1997). The proponents of the rational choice approach see voter behavior as determined by an evaluation of the costs and benefits anticipated in a potentially independent Québec as compared to the status quo. It is a prospective approach where respondents are invited to picture themselves in the future. For the Montréal School, for example, the possibility of economic problems in Québec following a victory of the yes side appears to be a dissuasive factor in the support for sovereignty. Members of the psycho-sociological school, on the other hand, see support for sovereignty as driven by grievances originating from a real or supposed inferiority of French-Quebeckers compared to the Anglo-Canadian majority, or even the Anglo-North American majority. According to the proponents of this theory, support for sovereignty will be higher as the feeling of inferiority, or the impression that the French language is

5 American Review of Canadian Studies 245 threatened, increases (Mendelsohn 2003). This perspective views the context as a relevant and substantial factor in the evolution of support for sovereignty. It states that the movement s assumed likelihood of success, combined with feelings of resentment toward Canadian federalism, are leading factors motivating Québec nationalism. Other studies focused instead on respondents demographic characteristics, notably in showing higher support for sovereignty among young (LeDuc 1977; Hamilton and Pinard 1982; Kornberg and Archer 1982; Gagné and Langlois 2002) and educated people (LeDuc 1977; Hamilton and Pinard 1982). However, recent studies have qualified these results (Durand 2008). It nevertheless remains that most studies approached the explanation of support for sovereignty at the individual level, based on the characteristics of respondents and their situations, their opinions, or their expectations of the future. In this study, we felt that a different approach was required one that focused on an analysis of determinants external to respondents. How did the question asked affect support? What does this tell us about the evolution of support and the decisions made regarding the proposals that were submitted to the referendums? How did the passing of time and the events that marked this period influence support? Others have previously examined the effects of question wording on change in support (Cloutier, Guay, and Latouche 1992; Pinard et al. 1997). Among other things, these studies show that questions mentioning the terms separation or independence generally led to weaker support than questions using the term sovereignty. Moreover, while they elicit stronger support in surveys, questions that combine sovereignty with an economic agreement with the rest of Canada sovereignty-association or sovereigntypartnership are also those that usually generate the highest rates of non-disclosers (Pinard et al. 1997). However, the methodological framework of these studies is generally weak (comparison of means, cross-tabulations, conclusions based on a small number of polls, little or no reference to the margin of error). Finally, most previous studies have focused on support for Québec sovereignty during referendum campaigns (see, among others, Drouilly 1997; Fox, Andersen, and Dubonnet 1999; Nadeau et al. 1999; Pinard and Hamilton 1984). For both referendums, studies examine only polls taken during the campaigns and sometimes a few weeks prior to the vote (Clarke and Kornberg 1996; LeDuc and Pammet 1995; Pinard and Hamilton 1984). Yet referendums are unusual contexts. Since the referendum question is known during the campaigns, pollsters tend to measure support for the question asked in the referendum, making it impossible to measure the effects of question wording during these periods. For all these reasons, this article seeks to paint a broad portrait of the changes in support for Québec sovereignty outside of referendum campaigns. It focuses on the effects of how poll questions are formulated and on the socio-political context in which the polls are carried out. We do not try to explain support at any given moment in history, but rather trace and explain how it changes over time. This article addresses three questions: (a) How does question wording affect measured support for sovereignty? (b) How does the context (time and events) influence measured support? (c) Does the effect of question wording vary over time and according to context? The data The database contains a total of 970 measures of support for Québec sovereignty, collected between 1962 and Some observations are incomplete and were consequently excluded from the database. 2 This leaves us with 857 usable observations.

6 246 F. Yale and C. Durand The entire period was broken down into shorter periods for analysis. The socio-political contexts are not totally comparable between periods and, consequently, question wordings, as well as related constitutional proposals that were polled, differ. We therefore divided the database into seven periods, based on the following considerations: The first period includes all data that preceded the first election of the Parti Québecois on November 15, 1976, with 41.4 percent of the vote. This was also a time when opinion polls started to be used somewhat more systematically in Québec. The period includes 12 valid polls, the first of which was conducted by Maurice Pinard with the Social Research Group in The second period covers late November 1976, just after the election of the Parti Québécois, to December 1979, when the official May 1980 referendum question was presented in the National Assembly. This relatively short period includes 68 valid polls, half of which were conducted in It leaves little room for longitudinal analysis. During most of this period except for a Conservative minority government headed by Joe Clark from June to December 1979 Pierre Elliott Trudeau, chief of the Canadian Liberal Party, was Prime Minister of Canada. The third period covers the 1980 referendum campaign, from when the referendum question was revealed to May 20, It includes 43 valid polls. The fourth period is the first post-referendum period, covering June 1980 to June This period has only 33 polls, including only one poll in November 1985 and none at all from 1986 to January 1988, a period when the sovereignty issue had all but disappeared from public discourse. Meanwhile, the Canadian Progressive Conservative Party (CPC) had been elected with Brian Mulroney as leader on September 4, 1984, and the Québec Liberal Party (QLP) had returned to power in Québec with Robert Bourassa as leader on Dec 2, 1985, with 56 percent of the vote. The fifth period extends from July 1989 (before the second election of the Québec Liberal Party on September 25 with 50 percent of the vote), to the eve of the 1995 referendum campaign in June. It includes multiple constitutional negotiations the Meech Lake and Charlottetown Accords and failures and the September 12, 1994, election, when the Parti Québécois took power with Jacques Parizeau as leader with 44.7 percent of the vote (barely more than the Québec Liberal Party at 44.4 percent). At the federal level, the Canadian Liberal Party returned to power on October 25, 1993, with Jean Chrétien as leader. This period includes 301 polls. The sixth period covers the 1995 referendum campaign. We positioned it between the three-way agreement on sovereignty-partnership on June 12, 1995, and the referendum itself held on October 30. It includes 74 valid polls. The seventh and last period (the post-1995-referendum period) includes all polls taken between November 1995 (right after the referendum) and January It includes 326 polls. This period is characterized by the sponsorship scandal. It includes the 1998 Québec election where the Parti Québecois was re-elected with Lucien Bouchard as the leader but with fewer votes than the Québec Liberal Party and the April 14, 2003 Québec election where the Québec Liberal Party returned to power with Jean Charest as leader. At the federal level, Jean Chrétien was re-elected in 1997 and 2000 and replaced by Paul Martin as the leader of the CLP in November The following governments were minority governments, either Liberal (Paul Martin, ) or Conservative (Stephen Harper, ). As stated, we excluded the two referendum campaigns from our analyses, first because they constitute special contexts that are difficult to compare to other so-called normal

7 American Review of Canadian Studies 247 periods, and secondly because they have been studied extensively. The pre-1976 period and the period were also excluded because they provide too few observations. The analysis therefore covers the pre-1980-referendum period, the pre-1995-referendum period, and the post-1995-referendum period, periods 2, 5 and 7 above, with a total of 695 observations. In this way we can examine how support for sovereignty changed during the run-ups to the referendums, along with the state of public opinion just before the campaigns kicked-off, as well as how it has changed since the last referendum. The measures In order to analyze the evolution of support for Québec sovereignty, one initial question to resolve is how to deal with the non-disclosers, sometimes called the undecided. 3 Depending on the period and the polling firm, the way non-disclosers are dealt with is either proportional allocation, non-proportional allocation with a larger portion attributed to the no side, or inferred by the polling firm based on answers to other questions. In order to standardize and compare like with like, we used a proportional allocation of nondisclosers, which amounts to the same thing as analyzing the proportion of yes votes over the sum of yes and no votes. Factors that may explain the evolution of support for sovereignty are related to either question wording or time and events. With respect to question wording, the first measure (voterint) indicates whether the question asked deals with voter intention or whether it is a general attitude question. Voter intention questions ask respondents if they will vote yes or no (or for or against ) in any future referendum. A reference to voter behavior here is essential. Questions of attitude measure opinions, and choices of response generally use alikertscale(e.g.,are you very much in favor, somewhat in favor, etc.). The second measure has to do with the type of constitutional proposal to which the question refers. The categorization takes its inspiration from that used in Pinard et al. (1997). The first category, called separation, is the most extreme of the proposals. It includes the word separation, the expressions complete political and economic independence or without constitutional ties to Canada, or refers to a Québec that is not a province of Canada. The second category is labeled independence. It includes questions with the single word independence or the expression independent country. The differentiating factor between these first two categories is mostly related to the clear absence of a relationship with the rest of Canada for separation. This distinction was suggested early on by Pinard and Hamilton (1977), when the terms independence and separation were often used interchangeably. The term separation has a negative connotation and has been widely used by federalist politicians in order to attack the project. On the other hand, the term independence has a more positive tone and was mainly used in the beginnings of the sovereigntist movement and remained used mostly by more radical elements on the sovereigntist side. The third category, sovereignty, refers to the expressions sovereignty, or sovereign state or country. A fourth category, sovereignty-partnership refers to the expressions sovereignty-association, sovereignty-partnership and other expressions that suggest an association between Québec and the rest of Canada. In addition, after numerous debates about the best question in the run-up to the 1980 referendum period, a number of polls measured support for a mandate to negotiate sovereigntyassociation ; however, this category exists only for that period. Finally, as we shall see below, for the pre-1980-referendum period, the proposals of separation and independence were eventually merged under the label extreme, given the similarity of the results for these categories and the statistical power required.

8 248 F. Yale and C. Durand Time related variables are of two types. They are either related to the passing of time itself (time variables) ortoeventsthatoccurredduringtheperiod(events variables). The base unit of time variables is the month. Hence, the variable month has a value of 1 for the first month of each period, and its value increases by 1 for each passing month. It allows estimation as to whether there is a linear increase or decrease in support for sovereignty with time. Non-linear change over time, i.e. evolutions that have a U or inverse-u form or a wave form with one trough and one crest, is also possible and was modeled using quadratic and cubic time effects (see Singer and Willett 2003). 4 A change in direction of support for sovereignty over time is, however, not the only possibility. After a significant event, support may undergo sudden jumps or drops. Subsequently, support may either maintain the same slope, or its rate of change may vary, indicating a change in trajectory. One can estimate the existence of a jump or drop in support by inserting a variable that takes the value of 0 before an event and a value of 1 after it. It is also possible to combine this instantaneous effect with linear, quadratic and cubic effects of time in the same way as it is done for the general analysis of the evolution. What events are likely to have influenced change in support for sovereignty? Broadly speaking they fall into two categories: elections, especially those in which there is a change in power; and events related to constitutional discussions, that is, the Meech Lake and Charlottetown accords, along with the sponsorship scandal and the Gomery Commission. During the period between November 1976 and December 1979, no such event was identified. For the period from June 1989 to June 1995, the events identified were the following: The failure of the Meech Lake Accord that occurred in June The question of a Meech effect has been studied by, among others, Cloutier et al. (1992) and Pinard et al. (1997). This event may have triggered grievances among French Québeckers. The rejection of the Charlottetown Accord in the October 1992 referendum. The election of the Parti Québécois on September 12, The hypothesis is that the election results, comforting to Quebeckers since a party dedicated to the defense of the French language and French Quebeckers interests was in power, could have led to a decrease in support for sovereignty due to a decrease in grievances along with an increase in the feeling of security. For the period between November 1995 and January 2008, the events identified are the following: The sponsorship scandal, officially starting in February 2004 when the Auditor General of Canada published her report. The election of the Québec Liberal Party on April 14, Much like the election of the Parti Québécois, albeit in the opposite direction, the election of the QLP has been hypothesized as increasing support for sovereignty, due to an increase in grievances and a decrease in the feeling of security. The publication of the Gomery Commission s report on November 1, The report confirmed the existence of a scandal in the way public funds had been spent. It attributed at least part of the responsibility to the Canadian Liberal Party and then-prime Minister Jean Chrétien. Finally, two methodological characteristics of the polls could have influenced the distribution of answers related to support for sovereignty. The first is sample size, since it could

9 American Review of Canadian Studies 249 theoretically affect how the results vary. The second is the proportion of non-disclosers. The mode of administration (i.e., in-person or telephone interview), could also have had an impact. However, there was too little variation in practice within periods only faceto-face for the first polls, and almost entirely by telephone starting in the 1980s for this possible effect to be tested. Analysis Hierarchical model Two sources of explanation are considered in this paper: survey characteristics and timerelated characteristics. Poll results are not completely independent from the context in which they are administered and effects related to measurement can vary over time. The objective was therefore to use a procedure that could estimate effects related to measure and time separately, as well as any possible change in effect related to measure and time together. The hierarchical model, also called the multilevel model takes into account the fact that polls (level 1) are nested within units of time (level 2). 5 Effects related to polls are analyzed at the poll level while effects related to time are analyzed at the time level. Procedure Three series of analyses one for each period were carried out, each with three steps, related to the three hypotheses. The first step estimated the effects of question wording. The second focused on the longitudinal dimension, i.e., effects related to time and events. The last step integrated both dimensions in a final, parsimonious model. In the first step, a model without any independent variable (Model Ø) was estimated. This model identified the variance attributable to each level. Next, the control variables and then the variables related to question wording (constitutional proposals and type of question) were entered. Sovereignty-association, considered the least extreme proposal, was used as the reference proposal, such that all other proposals were compared to it. Similarly, whether the question concerns voter intention was compared to whether it was a question of opinion. The second step focused on change over time. In order to examine the overall evolution in popular support, a level-2 model that ignored question wording was estimated. Variables related to evolution with time along with those representing the possible impact of each event were first tested individually. The parsimonious model retained the significant effects, which led to an integrated average model for change over time. Finally, in the last step, we estimated the evolution of support for the different question wordings and types. Each of the variables related to the constitutional proposals except sovereignty-association, the reference category were integrated into the model described above in order to see whether the evolution of support for these proposals differed on average from the evolution of support for the reference proposal. Results Pre-1980 referendum period The first model (Model Ø) which contains no predictors estimates the average support for sovereignty during the period and the distribution of variance between the two levels of analysis. This model s estimates show that nearly all the variance (97 percent) occurs at

10 250 F. Yale and C. Durand level 1, the level pertaining to measure. Variation thus occurs within months and can only be explained by survey and question characteristics and not by differences between months. This suggests a stability of public opinion over time during this period. Thus, the next steps aim at explaining level 1 variance. The groups of variables were entered step by step, i.e., first, the two control variables (sample size and proportion of non-disclosers), then the variable measuring the fact that avoterintentionquestionwasasked(voterint) andfinally,thevariablesrelatedtothe constitutional proposals. The control variables both appeared not significant and were therefore not included in subsequent models. The final parsimonious model is presented in Table 1. In addition to voterint, itcontainsthevariablesmandate (for mandate to negotiate) and extreme. 6 Sovereignty was not part of the political discourse during that period and was not included in the questions asked by pollsters. The voterint variable is not significant. The coefficients for the two proposals are significant, strong and in the direction predicted by theory (mandate: 14.01, P<0.001; extreme: , P<0.001). With respect to support for a proposal on sovereignty-association, questions dealing with the mandate to negotiate collected about 14 percentage points more support, while those dealing with separation or independence collected on average 20 points less support. Compared with Model Ø, this model explained 89 percent of the level-1 variance in the support for sovereignty. On the other hand, the level-2 variance component, which was almost nil in Model Ø, increased and became significant (10.10, P<0.001), which can be explained by a relationship between the type of proposal and time, indicating that testing time-related factors is relevant. However, only 21 time units (months) during this period had at least one measure of opinion. In theory, it is recommended to have at least 40 level-2 units to carry out multi-level analysis. Thus, the results should be interpreted with great caution. Table 1. Average impact of question wording on support for sovereignty Fixed effects Intercept (1.51) (1.05) (1.57) Voterint n.s (0.76) 4.04 (1.35) Sovereignty 7.63 (1.09) 6.75 (0.87) Independence 13.46(0.90) 8.95 (1.31) Separation (1.08) (1.52) Mandate (0.85) Extreme (1.67) Size n.s. n.s. n.s. Non-disclosers n.s. n.s (0.0941) Variance component Level-1 R (%) Level-2 Intercept (%) Deviation Parameters DL Note: P < 0.05; P < 0.01; P < 0.001; n.s. not significant. The variable was tested in one previous model and removed from the model.

11 American Review of Canadian Studies 251 The analysis showed no significant effect of time variables. The results suggest that either there was no significant movement in opinion between November 1976 and December 1979, or that the data are not sufficient to identify such movement. We conclude that, on the eve of the 1980 referendum campaign, surveys placed support for sovereignty-association at 39.8 percent after proportionally distributing undecided voters. This support rose to 53.8 percent ( ) when the question dealt with a mandate to negotiate such sovereignty and dropped to 19.5 percent ( ) when the question dealt with separation or independence. It should be pointed out that with experience, both pollsters and researchers subsequently concluded that a proportional distribution of undecided voters overestimated support for sovereignty and that instead, two-thirds or even three-quarters of them should be attributed to the no side. During the few months before the 1980 referendum, polls continued to show strong support for the official question pertaining to a mandate to negotiate sovereigntyassociation (48 percent on average). However, on May 20, 1980 a solid majority of Québeckers, 59.6 percent, voted against that proposal. Pinard and Hamilton (1984) suggested that the discrepancy between measured voting intention and the final vote may have been due to the fact that voters cast their vote according to what they saw as the real issue, i.e., Québec sovereignty, rather than the specific question asked. The following years were quiet with regard to the Québec sovereignty issue. There were also too few polls to conduct any analysis. It is only at the end of the 1980s, with the upcoming failure of the Meech Lake Accord, that focus was brought back to the issue. The pre-1995 referendum period During the pre-1995 referendum period, from 1989 to 1995, Model Ø suggests that 62 percent of the total variance in support for sovereignty could be explained by factors related to question wording, with the remaining 38 percent related to the level of time. The final model, presented in Table 1, shows that control variables were not significant. Questions pertaining to voter intention collected an average of 3.2 fewer percentage points than questions pertaining to attitude. Compared to questions pertaining to sovereignty-partnership, those on separation collected 16.8 points less support, those on independence, 13.4 points less, and those on sovereignty, 7.6 points less. The average effects of the separation and independence proposals were not significantly different from each other. Compared with Model Ø, this model explained 57 percent of the measurement related variance and 29 percent of the time-related variance. Finally, 51 percent of the total variance was now explainable by time-related factors. This brings us to the second step of the analysis. There were 58 valid time units for this period. Tested individually, each event considered for this period had a significant effect on how public opinion changed. However, since some events are concurrent or correlate with the evolution with time, when the variables were entered jointly in a single model, only three effects remained significant: one linear effect of time, the failure of the Meech Lake Accord and the Referendum on the Charlottetown Accord. The election of the PQ in September 1994 did not appear significant. The final step integrated the measurement-related variables to the model of change over time, i.e., the longitudinal model above was estimated for each constitutional proposal. These models show that all three time-related variables had a significant impact on the evolution of support for sovereignty, but that these effects differ from one constitutional proposal to another. Taking into account the coefficients specific to the reference model, we obtained a longitudinal model for each of the proposals. The resulting models, taking

12 252 F. Yale and C. Durand into account the average effect of asking a question on voter intention (-1.74; P=0.01), are presented in Table 2 and reproduced graphically in Figure 1. The beginning of this period was marked by an increase in support for all proposals though this increase was rather weak for separation until the final defeat of the Meech Lake Accord in June The impact of this failure was not measured until September due a lack of polls over the summer, hence the delay effect observed. There was likely a stabilization of support after Meech Lake, followed by a drop in support for the different proposals. The rejection of the Charlottetown Accord by the population slowed down this decrease in support for the sovereignty-partnership proposal. For the three other proposals, the change in slope was not significant. Eventually, support for these latter proposals converged towards the same level of support, below sovereignty-partnership. The final model explains 65 percent of the variance at the measurement level and 88 percent of the variation over time. We may therefore conclude that during the pre-1995-referendum period, support for Québec sovereignty, regardless of question wording, was in a gradual decline starting from Table 2. Final models: evolution of average support for sovereignty for and Sovereignty-partnership Sovereignty Independence Separation intercept month meech charlot intercept month month sponsorship gomery P < 0.05; P < 0.01; P < % support sov-partership sovereignty independence separation Jul 89 June 90 : Meech Sept 91 Oct 92 : Charlottetown month/year Jan 94 Sept 94 : PQ elected Jan 96 Figure 1. Evolution of support for sovereignty

13 American Review of Canadian Studies 253 the official failure of the Meech Lake Accord. Moreover, as the referendum approached, support for the sovereignty-partnership proposal increasingly stood out from support for the other proposals. By the end of the period, on the eve of the 1995 referendum campaign, intention to vote for sovereignty accompanied by a partnership with the rest of Canada had support of nearly 52 percent 7 of the population after proportional distribution of undecided voters, 4 percentage points more than at the start of the period. The other, more extreme, proposals separation, independence and sovereignty without mention of an association had only slightly more than 40 percent support, about the same as at the start of the period. The 1995 referendum campaign began in June with the three-way agreement between Parti Québécois, Action Démocratique du Québec and Bloc Québécois and the unveiling of the official referendum question. From this moment on, there was quite a lot of movement in support for sovereignty-partnership. First, from the second half of June to early October, polls showed a slight decline in support for the yes side (from 51 percent in late June to 48 percent in early October). However, with the arrival of the charismatic sovereignist leader Lucien Bouchard at the forefront of the movement, the yes side gained more support in public opinion polls (Durand 2008). On the eve of the referendum, support for both sides was more or less even. Some of the published polls predicted a win for the yes side (Createc and SOM polls, on October 25, with 53 percent for the yes side). On October 30, 94 percent of eligible voters went to vote. This time, pollsters had it right. The no side eventually got the support of 50.6 percent of Quebeckers, barely more than the yes side. Post-1995 referendum period For the period, the distribution of Model Ø variance indicated that 89 percent of total variance was measure-related, while time-related factors explained just 11 percent, suggesting that variation over time was clearly less than during the previous period. Estimates for question-wording effects are presented in Table 1. Voter intention questions gathered on average 4 percentage points fewer than attitude questions. When compared with the reference proposal sovereignty-partnership questions on separation collected 11.4 points less support, those on independence 8.9 points less and those on sovereignty 6.8 points less. The average effects of the independence and sovereignty proposals during this period did not differ significantly. The same went for the effects of the independence and separation proposals. Only sovereignty and separation differed significantly. Compared with Model Ø, this model explained 50 percent of the measurement-related variance. The inclusion of these variables resulted in an increase of the time-related variance, which may mean that the different proposals evolved differently over time. As with the previous period, we first identified an average longitudinal model, and we then estimated a model for each of the proposals (see Table 2). The voter intention question still had a significant mean effect ( 4.12; P = 0.001). Figure 2 illustrates the evolution of voter intention for the four constitutional proposals. Support for all proposals fell substantially at the start of the period. Support stabilized in late 2001 and the election of QLP in April 2003 did not appear to significantly influence the support for sovereignty. However, the sponsorship scandal, which emerged in the spring of 2004, appears to have caused a sudden increase in support for all proposals. After a plateau that lasted during the entire period that the Gomery Commission was investigating the scandal, a drop in support for two of the proposals can be seen following the publication

14 254 F. Yale and C. Durand % support sov-partership sovereignty independence separation Figure Nov 95 June 97 June 99 June 01 month/year Evolution of support for sovereignty April 03 : PLQ elected Mar 04 : sponsorship Jan 05 Nov 05 : Gomery Report Nov 06 of Judge Gomery s report in November This effect could not be tested for the independence and separation proposals due to a lack of data. As with the failure of the Meech Lake Accord, the publication of Judge Gomery s report appeared to conclude instead of start a period of increase in support for sovereignty. This model explained a total of 52 percent of the measure-related variance and 55 percent of the time-related variance. At the end of the period, sovereignty-partnership i.e., the subject of the question asked in the 1995 referendum (the last poll using this question in our database is the CROP poll of April 2007) collected 42 percent of voter intentions after a proportional distribution of undecided voters, and sovereignty without partnership (the last poll being the CROP poll of January 2008) collected 39 percent support. At the beginning of the period, as illustrated in Figure 2, these two proposals collected between 45 percent and 55 percent support. The event that had the most substantial impact the sponsorship scandal did not cause support to rise to that level for any length of time during the period. Discussion 20 In response to the three initial research questions, we can conclude that the type of question (voter intention rather than overall attitude) had an effect on the results obtained in the polls. First, voter intention questions appeared to elicit weaker support than attitude questions. Second, the pre-1995-referendum period showed fairly strong average effects for constitutional proposal, though such effects were less substantial (and not all significant) for the following period. Regarding the second hypothesis, we observed that time and certain events could explain a non-negligible portion of the variation in measured support. The effect of the failure of the Meech Lake Accord is a perfect illustration: this event corresponded to a significant change in how support for sovereignty varied over time. After an increase in support of nearly 20 points in the run-up to the Meech deadline, the confirmation of its failure appears to have led to a growing public apathy that manifested itself in a slowed

15 American Review of Canadian Studies 255 increase and eventually a drop in support. However, this decline was not as significant for sovereignty combined with an association or partnership with the rest of Canada. While this observation may seem counterintuitive, the phenomenon has also been noticed and discussed in previous work (Cloutier et al. 1992; Pinard et al. 1997). The official failure of the Meech Lake Accord in June 1990 was not a surprise, and popular dissatisfaction with the deadlock in constitutional negotiations had already begun to be felt several months before. Support might well have remained high, but it did not. One possible explanation is the concept of sovereignty threat introduced by Jean-Herman Guay (2004). If, as Guay suggests, some voters who say they support sovereignty do so only to demonstrate their discontent with the government, as was the case with the Meech Lake Accord, its failure was an end in itself. Therefore, the role played by public opinion in holding politicians in check during these constitutional negotiations was no longer necessary. It is also possible that support for sovereignty had already reached its peak with the upcoming failure of the Meech Lake Accord. Hence, from this point on, support could only go down. According to such an hypothesis, it would not be the actual failure of the agreement, but the simple passing of time that would have caused support for sovereignty to go down. These explanations could be complementary as well. Other events also had an impact, though sometimes a rather weak one. The failure of the Charlottetown Accord in 1992 and the election of the Parti Québécois in 1994 did not seem of great consequence, aside from the fact that the failure of Charlottetown corresponded to a slowdown of the fall in support for sovereignty-partnership. However, this slowdown is not necessarily due to Charlottetown. Support may as well have simply leveled out, meaning that the various proposals all had a partisan basis and therefore a minimum threshold of support. During the post-1995-referendum period, the emergence of the sponsorship scandal in 2004 appears to have influenced support for all proposals, while the election of the QLP showed no effect when tested in conjunction with other events. Finally, the Gomery Commission affected support for the sovereignty and sovereignty-partnership proposals. Though it is not possible to assess an independent effect of the publication of the report since it was concurrent with the beginning of the federal electoral campaign, this event did correspond to a decline in support for sovereignty. This suggests an interpretation similar to that used to explain the effect of the Meech Lake Accord failure. The conclusions for the last hypothesis how change in support varies according to question wording differ from one period to the next. It is clear that between 1989 and 1995, support for the different proposals did not change in parallel. There is evidence to suggest a trend towards convergence for all proposals other than sovereignty-partnership. Could this be a specific effect caused by the impending referendum? Indeed, following the referendum, from 1995 to 2008, constitutional proposals appeared to differ less than during the preceding period, suggesting there was less confusion due to question wording. Besides, the evolution of support appears to be parallel for the three main proposals, contrary to what had appeared in the pre-1995-referendum period, i.e., a different evolution for sovereignty-partnership. Thesedifferencesinthemeaneffectofconstitutionalproposals between periods are significant for all proposals but sovereignty. Conclusion This meta-analysis confirms previously observed results and opens up new avenues of reflection. The study stands out in terms of its methodology. The use of multi-level